Every Language Model Has a Forgery-Resistant Signature

TL;DR

We show that all language models impose elliptical constraints on their outputs, which can be used as a hard-to-fake signature to identify a model from its outputs.

Opening excerpt from the authors’ abstract. source